Beta Docs

Documentation

Everything you need to get started with DeepNode.

Quick Start

Get up and running with DeepNode in under 5 minutes. This guide walks you through installation, authentication, and deploying your first model.

Prerequisites

  • Python 3.9 or later
  • A DeepNode account (request access)
  • A trained model (any major ML framework)

1. Install the CLI

$ pip install deepnode-cli

2. Authenticate

$ deepnode auth login

This opens your browser for OAuth authentication. Your API key is stored locally in ~/.deepnode/credentials.

3. Deploy a model

$ deepnode deploy --model ./my_model --gpu nova --replicas 3

DeepNode automatically detects your framework, builds a container, and deploys to the nearest region. You'll receive an inference endpoint within minutes.

Installation

pip (recommended)

$ pip install deepnode-cli

Shell script (macOS / Linux)

$ curl -fsSL https://get.deepnode.cc | sh

Container

$ deepnode pull cli:latest
$ deepnode run cli --version

Verify installation

$ deepnode --version
deepnode-cli 0.14.2

CLI Reference

The DeepNode CLI provides a complete interface for managing your infrastructure from the terminal.

Global flags

--project <id> Override the active project
--region <region> Target region (eu-west, us-east, ap-south)
--output <format> Output format: table, json, yaml
--verbose Enable debug logging

deepnode deploy

Deploy a model to production infrastructure.

$ deepnode deploy --model <path> [options]

Options:
--gpu <type> GPU type: spark, pulse, nova, apex (default: pulse)
--replicas <n> Number of replicas (default: 1)
--min-replicas <n> Minimum replicas for autoscaling
--max-replicas <n> Maximum replicas for autoscaling
--region <region> Deployment region
--env <key=val> Environment variables (repeatable)

deepnode train

Launch a distributed training job.

$ deepnode train --config <file> [options]

Options:
--cluster <name> Target cluster
--nodes <n> Number of nodes (default: 1)
--gpu-per-node <n> GPUs per node (default: all available)
--priority <level> Job priority: low, normal, high
--tag <name> Tag for experiment tracking

deepnode serve

Serve a registered model version.

$ deepnode serve --model <name:version> [options]

Options:
--autoscale Enable autoscaling
--timeout <ms> Request timeout (default: 30000)
--batch-size <n> Maximum batch size for inference

deepnode status

View status of deployments and training jobs.

$ deepnode status [deployment-id]
$ deepnode status --all

deepnode logs

Stream logs from a deployment or training job.

$ deepnode logs <deployment-id> --follow --tail 100

API Reference

The DeepNode REST API is available at https://api.deepnode.cc/v1. All requests require an API key passed via the Authorization header.

Authentication

Authorization: Bearer dn_live_xxxxxxxxxxxxxxxxxxxx

List deployments

GET /v1/deployments

Response:
{
"deployments": [
{
"id": "dep_abc123",
"model": "my-model",
"version": "2.4.1",
"status": "running",
"region": "eu-west",
"replicas": 3,
"endpoint": "https://my-model.serve.deepnode.cc"
}
]
}

Create deployment

POST /v1/deployments

Body:
{
"model": "my-model:2.4.1",
"gpu": "nova",
"replicas": 3,
"region": "eu-west",
"autoscale": {
"min": 2,
"max": 10,
"target_gpu_utilization": 0.7
}
}

Run inference

POST /v1/inference/<deployment-id>

Body:
{
"inputs": { ... }
}

Response:
{
"outputs": { ... },
"latency_ms": 42,
"model_version": "2.4.1"
}

Get metrics

GET /v1/deployments/<id>/metrics?period=24h

Response:
{
"requests": 145230,
"avg_latency_ms": 47,
"p99_latency_ms": 89,
"gpu_utilization": 0.68,
"error_rate": 0.001
}

Deployment

Regions

DeepNode is available in the following regions:

  • eu-west — Northern Europe
  • eu-central — Frankfurt, Germany
  • us-east — Virginia, USA
  • us-west — Oregon, USA
  • ap-south — Singapore

GPU types

  • Spark — Budget-friendly inference (16 GB VRAM)
  • Pulse — Balanced training/inference (24 GB VRAM)
  • Nova — High-performance training (80 GB VRAM)
  • Apex — Maximum throughput (80 GB HBM3) — Enterprise only

Autoscaling

Enable autoscaling to automatically adjust replicas based on GPU utilization or request queue depth.

$ deepnode deploy --model ./my_model \
--gpu nova \
--min-replicas 2 \
--max-replicas 10 \
--autoscale

Environment variables

Pass environment variables to your deployment:

$ deepnode deploy --model ./my_model \
--env MODEL_PRECISION=fp16 \
--env MAX_BATCH_SIZE=32

FAQ

What frameworks are supported?

DeepNode supports all major ML frameworks. Framework detection is automatic — just point us at your model directory.

How does billing work during the beta?

During the public beta, Starter tier is completely free (100 GPU hours/month). Pro and Enterprise plans are billed monthly based on usage. See pricing for details.

Can I use my own container images?

Yes. While DeepNode auto-detects frameworks and builds containers for you, you can also bring a custom container image. Specify it with --image during deployment.

Is my data encrypted?

All data is encrypted at rest (AES-256) and in transit (TLS 1.3). Enterprise customers can additionally enable VPC peering for private network connectivity.

How do I get support?

Starter users get community support via our forum. Pro users get priority email support. Enterprise customers receive 24/7 dedicated support with a guaranteed SLA. Reach us at hello@deepnode.cc.

Where is DeepNode hosted?

Our primary infrastructure runs on Tier III+ data centers across Europe and North America.